Parquet writer column_size() should return a size_t

[etseidl]

Description

Fixes #12867.

Bug introduced in #12685. A calculation of total bytes in a column was returned in a 32-bit size_type rather than 64-bit size_t leading to overflow for tables with many millions of rows.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[rapids-bot]

Pull requests from external contributors require approval from a rapidsai organization member with write or admin permissions before CI can begin.

[etseidl]

Wanted to get this out fast for testing. I still have to add a unit test.

[vuule]

/ok to test

[vuule]

Thanks for the quick PR @etseidl !

[karthikeyann]

LGTM 👍

[galipremsagar]

Thanks @etseidl, I'll test this out today and get back

[galipremsagar]

Thanks for the quick fix @etseidl ! Verified that this PR fixes the issue #12867

[galipremsagar]

/merge

[etseidl]

Thanks for the quick fix @etseidl ! Verified that this PR fixes the issue #12867

Sorry for the bug @galipremsagar ☹️

[etseidl]

Since CI hates me, should I push the unit test I wrote for this? @vuule

[vuule]

Since CI hates me, should I push the unit test I wrote for this? @vuule

How big/slow is the test? My concern is that it requires a lot of device memory to execute, so it would not run on all GPUs.

[etseidl]

How big/slow is the test? My concern is that it requires a lot of device memory to execute, so it would not run on all GPUs.

Well, it writes 300M int64's, so 2.4GB. Takes about 1.5sec on my A6000. nsys says 24GB memory used 😮

[vuule]

nsys says 24GB memory used 😮

FWIW, that's the size of the pool, so peak memory use could be a lot lower. Still, I have an 8GB GPU and it's unlikely that it's enough for a 2.4GB file. So let's leave this one without a test, the fix looks very safe anyway.

[vuule]

/merge

[vuule]

/ok to test

[galipremsagar]

@vuule CI currently is blocked by #12874, I just canceled the jobs of this PR to free up machines for 12874 to finish quicker

[vuule]

@vuule CI currently is blocked by #12874, I just canceled the jobs of this PR to free up machines for 12874 to finish quicker

Thanks. I saw CI passing on another PR (probably ran at a different time) and got the wrong idea :)

[galipremsagar]

@vuule CI currently is blocked by #12874, I just canceled the jobs of this PR to free up machines for 12874 to finish quicker

Thanks. I saw CI passing on another PR (probably ran at a different time) and got the wrong idea :)

CI is now ready. Thanks for bearing with #12874 😉

[galipremsagar]

/ok to test

[galipremsagar]

/ok to test